FSD Beta Videos (and questions for FSD Beta drivers)

[gearchruncher]

Once perception is solid, the rest is just a deep learning issue. Labeled data+compute+time = fast improvements. Dojo will help accelerate the trend. Personal guess: L5 by 2023.

with L5 being achieved, which is indeed a monumental task that might not happen in our lifetime (you never know, it hasn't happened yet. Completely uncharted territory).

I'm guessing TSLA has a shot at L5 by 2023 but of course I know that's a long shot and more likely it's much later than that. Not that I'm backing down, just saying it's easy to be Negative Nancy but not adding any value.

Your personal guess is 2023, but it's a monumental task that might not happen in the next 50 years either. Sounds like something a lawyer would write. Thanks for your helpful, data driven opinion!

My bet is in Tesla stock.

Oh wait.. There's the source of your opinion

I'm guesstimating just as much as you.

No, I'm using data. FSD has been "solved" since 2015, yet it still hasn't shipped. "FSD" V8 came out 9 months ago, and this is the progress we have so far between V8 and V9. There is no extrapolation of anything humans have ever done that gets you to L5 in 17 months from where we are now.

 

[gearchruncher]

you are gonna have to offer up better odds. 5000k today for 10000k tomorrow isn’t as attractive as you say it is because I can probably put that same amount of money in Tesla stock and get the same if not greater return...

My odds were against people saying things will be true in a month or two. 100% guaranteed return in 2 months? It's an amazing deal.

Unless maybe you do think it's could actually be 24 months... Then, yeah, that's the point. You're not going to take it.

These posters were so sure we were L2 better than a human in 1 month or L5 in 17 months, but when odds were offered that paid them a lot if that was true, but lost them money if it wasn't, they suddenly were less sure.

I'll still take anyone up on selling them FSD today for $5K, and will buy them FSD at whatever price it is on the day Tesla releases what they described in 2016. Who doesn't want FSD for half off?

 

[ZeApelido]

Your idea that real world data is the solution and simulation is worthless and overhyped. Clearly the opposite is happening as companies (waymo) who invested more in SOTA sim is thriving while Tesla is struggling to break the 10 mile safety barrier.

NO

The "solution" is Real World Data (A) x Simulation (B). I didn't say simulation is worthless, this seems to be a common confusion among laypersons.

A bigger "A" amplifies the effect of "B". This has nothing to do with Tesla. I look at it from a holistic and conservative approach - conservatively for me to expect someone's self driving to work well anywhere in the U.S. will require sufficient real data collect from all over the U.S. That Waymo is not letting their vehicles go and show off their self driving tests in any city in the U.S. supports the notion that even Waymo truly understands this.

Mark me, Waymo is the leader. If Waymo had their sensors on 100,000 cars deployed all over the U.S., I would have no quarrel with their approach or statements. I have never said Tesla will beat Waymo, that is just your projection to try to simplify my perspective.

 

[rxlawdude]

Whole Mars almost gets into a minor accident here, lol:

View attachment 683363

https://twitter.com/x/status/1414044620388921346

You do realize the video was sped up, making this look much worse than in reality, right?

 

[rxlawdude]

What about this video where V9 fails miserably and is way worse than Waymo?

This why I warn against making judgments based on videos. It is too easy to cherry pick. Yes, you can find videos where V9 appears to perform better or appears to perform as good as Waymo in certain cases but there are also videos of V9 performing way worse.

Except that FSD at this stage does not aspire to be driverless. You seem to ignore that. Then again, that video where Waymo gets absolutely stuck due to construction zone cones is a classic. That's designed to be driverless.

Pot, see kettle.

 

[Matias]

Except that FSD at this stage does not aspire to be driverless.

It doesn’t?!

 

[rxlawdude]

It doesn’t?!

What part of "at this time" don't you understand? That's like denying that Waymo aspires to work in non-geofenced areas at this time.

 

[Bladerskb]

Except that FSD at this stage does not aspire to be driverless. You seem to ignore that. Then again, that video where Waymo gets absolutely stuck due to construction zone cones is a classic. That's designed to be driverless.

Pot, see kettle.

huh? this makes no sense. One car breaks down every couple miles and need instant takeover to avoid an accident (OMG ITS AMAZING. LIGHT YEARS AHEAD).
The other car is completely driverless, has given tens of thousands of miles driverless and can go 30k miles without safety disengagement in CA. (OMG IT SUCKS, REMEMBER THAT TIME IT STOP FOR CONES IN THE ROAD AND DIDN'T CRASH. WHAT A JOKE. YEARS BEHIND!)

 

[qdeathstar]

My odds were against people saying things will be true in a month or two. 100% guaranteed return in 2 months? It's an amazing deal.

Unless maybe you do think it's could actually be 24 months... Then, yeah, that's the point. You're not going to take it.

These posters were so sure we were L2 better than a human in 1 month or L5 in 17 months, but when odds were offered that paid them a lot if that was true, but lost them money if it wasn't, they suddenly were less sure.

I'll still take anyone up on selling them FSD today for $5K, and will buy them FSD at whatever price it is on the day Tesla releases what they described in 2016. Who doesn't want FSD for half off?

ok, I was thinking more over a 5 year horizon.

 

[qdeathstar]

FSD Beta V9 only approaches Waymo in very simple driving tasks. It does not approach Waymo in more complicated tasks. V9 is L2 "door to door", it is not autonomous driving. Waymo is actual L4 autonomous driving. And yes, V9 could probably handle easy routes in Chandler. Waymo's 5th Gen can handle more complicated environments than Chandler. Waymo's 5th Gen FSD is light-years ahead of V9.

where is it deployed in more complicated environments that allows you to make trust assertion?

 

[Bladerskb]

You do realize the video was sped up, making this look much worse than in reality, right?

You do realize that speeding up hides the gravity of the situation not mask it?

 

[Bladerskb]

NO

The "solution" is Real World Data (A) x Simulation (B). I didn't say simulation is worthless, this seems to be a common confusion among laypersons.

There's no confusion. The problem is you unable to grasp simple simulation concepts. Just weeks ago you were saying data augmentation is the same as simulation.

A bigger "A" amplifies the effect of "B".

This is something that you seemingly are unable to grasp. The industry is not going the other way. They are not trying to get more real world data.
The opposite is completely true. They are improving B to the point that they need less A. The more realistic and complex B is, the less A is required and the more they can improve their software exponentially using more B.

You simply can't grasp that, that's why you run around saying " Simulations are not creating weird cases that may only happen in Fargo, South Dakota because Waymo engineers haven't come across those weird cases before." or "simulation doesn't replace real world edge cases".

Sim does exactly that. Not only is the AV industry moving towards more complex and realistic simulation as seen with Waymo's Simulation Cities. The AI industry at large is doing so as well.

Your sim can get so good that you only need a tiny amount of real world data as a bootstrap or none at all.

"Today, a team of researchers from Facebook AI, UC Berkeley, and Carnegie Mellon University’s School of Computer Science are announcing Rapid Motor Adaptation (RMA), a breakthrough in artificial intelligence that enables legged robots to adapt intelligently in real time to challenging, unfamiliar new terrain and circumstances. RMA uses a novel combination of two policies, both learned entirely in simulation"

AI now enables robots to adapt rapidly to changing real world conditions

Researchers from Facebook AI, @berkeley_ai and @SCSatCMU have developed AI that can enable a legged robot or other machines to adapt in fractions of a second to changing conditions in the real world.

ai.facebook.com

https://twitter.com/x/status/1413490134129922050

This has nothing to do with Tesla. I look at it from a holistic and conservative approach - conservatively for me to expect someone's self driving to work well anywhere in the U.S. will require sufficient real data collect from all over the U.S. That Waymo is not letting their vehicles go and show off their self driving tests in any city in the U.S. supports the notion that even Waymo truly understands this.

Huawei Autopilot which will be released in Q4 on the 2021 Arcfox Alpha S works anywhere in china.
Mobileye's Supervision which will be released on the 2021 Zeekr 001 works anywhere in the world.

They are all doing that without 100k or even 1k cars roaming around. You're simply wrong from every angle. The industry is doing more sim and less real world.
Not the other way around.

 

[AndreP]

L5 autonomy would require the ability to operate anywhere at any time wouldn't it?

If so, there is no way it's happening by 2023 and I'd say surely not this decade. Beta testing in cities like San Fran or Seattle might be decent trials for hectic driving in the West, but those are also among the most highly developed cities in the world. Roads are well maintained and traffic laws/conventions are mostly adhered to, which can't be said for many other large cities around the world much less everything in between those cities.

Will be quite surprised if I see a true L5 car, working all over the world and in all weather conditions, within my lifetime

 

[mikes_fsd]

wish some technical people had the beta, not just youtubers. James Douma, George Hotz, Green etc.

This, I will have to disagree on.

Non-technical people is going to be so raw as it can get without trying to over complicate it.

 

[powertoold]

You do realize the video was sped up, making this look much worse than in reality, right?

Yup I took that into acount. V9 performance in that situation was still incredible, as it saw all the parked cars and also the moving car but it should have slowed down or stopped.

 

[mikes_fsd]

That ride was epic fail.

That ride was a test that is needed by Tesla.

It was a perfect example why the closed beta is needed.

 

[diplomat33]

That's why I said some of these "drives" are definitely approaching Waymo performance. If we were to put V9 in Chandler doing the same routes (mostly protected turns and detours through easy streets), I'm sure it'd be similar performance. Waymos do well when the environment is simple and predictable (don't go showing me some video provided by Waymo to show it does well, lol. I just watched a JJRicks video where Waymo got stuck again in a simple situation).

It's unsurprising that V9 fails in SF, mostly because of its poor lane choices / confusion and poor lane changing speed.

It is also important to note that while V9 and Waymo might look the same in capability, they are quite different. It is the difference between a driver assist and an autonomous driving system. FSD Beta V9 requires stalk confirmation to make turns into high speed traffic. And we see the driver check for cross traffic and then confirm the turn. So the driver is telling FSD Beta when it is safe to make the turns. So FSD Beta is getting help from the human driver as to WHEN to make the turn which is key. Waymo has no driver. So when Waymo makes the same turn, it is truly the Waymo doing the entire DDT, not just steering, but also checking for cross traffic and making the key decision on its own about when it is safe to make the turn. Big difference!

 

[powertoold]

It is also important to note that while V9 and Waymo might look the same in capability, they are quite different. It is the difference between a driver assist and an autonomous driving system. FSD Beta V9 requires stalk confirmation to make turns into high speed traffic. And we see the driver check for cross traffic and then confirm the turn. So the driver is telling FSD Beta when it is safe to make the turns. So FSD Beta is getting help from the human driver as to WHEN to make the turn which is key. Waymo has no driver. So when Waymo makes the same turn, it is truly the Waymo doing the entire DDT, not just steering, but also checking for cross traffic and making the key decision on its own about when it is safe to make the turn. Big difference!

Totally agree with you. My opinion is that V9 would be able to perform just as well in 95% of the drives (same routes) Waymo does in JJRick's videos. Waymo is still better in a lot of ways, especially in parking lots. But we also have to consider that Waymo is driving in fully mapped environments, whereas V9 is "seeing things for the first time" and is a way more generalized approach. This means that at every second, V9 is considering all the possibilities of all USA roads. Even driving straight with V9 is not trivial, as the computer doesn't make any assumptions. It has to drive based on what it thinks it sees and has no backup route to follow.

 

[superblast]

It is also important to note that while V9 and Waymo might look the same in capability, they are quite different. It is the difference between a driver assist and an autonomous driving system. FSD Beta V9 requires stalk confirmation to make turns into high speed traffic. And we see the driver check for cross traffic and then confirm the turn. So the driver is telling FSD Beta when it is safe to make the turns. So FSD Beta is getting help from the human driver as to WHEN to make the turn which is key. Waymo has no driver. So when Waymo makes the same turn, it is truly the Waymo doing the entire DDT, not just steering, but also checking for cross traffic and making the key decision on its own about when it is safe to make the turn. Big difference!

Are there any good videos of Waymo making an unprotected left from a stop sign into higher speed traffic like the Chuck video intersection?

 

[diplomat33]

Are there any good videos of Waymo making an unprotected left from a stop sign into higher speed traffic like the Chuck video intersection?

Check JJ's video repository. The videos are annotated. There are lots of unprotected turns. I am sure you can find an unprotected left turn from a stop sign.

JJRicks.com - Videos

Find the full playlist of videos here or at jrj.pw/waymo for quick access.

www.jjricks.com